ActiveRecord Hstore bug: can't update a key in the hash

[joevandyk]

joevandyk@f2318b2 shows a failing test.

  def test_updating_key                                                                             
    x = Hstore.create! :tags => { "key1" => "old value 1", "key2" => "old value 2" }                
    x.reload                                                                                        
    assert_equal "old value 1", x.tags["key1"]                                                      

    # Nothing gets saved/updated here.
    x.tags["key1"] = "new"                                                                          
    x.save!                                                                                         

    assert_equal "new", x.reload.tags["key1"]                                                       
    assert_equal "old value 2",   x.reload.tags["key2"]                                             
  end

  1) Failure:
test_updating_key(PostgresqlHstoreTest) [cases/adapters/postgresql/hstore_test.rb:55]:
Expected: "new"
  Actual: "old value 1"

[kennyj]

I can reproduce this issue.
It seems that any query weren't executed when executing save!
I guess:
x.tags method returns a Hash object, and we can't track dirty flag.

[gertig]

I can also reproduce this issue.

[al2o3cr]

A quick workaround would be to call (in your example) x.tags_will_change! before setting the new value.

Historically, this has always been a problem with serialize columns, and AR's approach in that case is to always save serialized columns. HStore columns might need the same treatment.

[tenderlove]

@al2o3cr ya, I've been looking at this. I'd like to treat the hstore columns as serialized columns, but it's going to take a little refactoring.

[parndt]

@tenderlove has there been any refactoring to date that would have helped this?

[neerajsingh0101]

Here is another version and this one does not break the API. It is based on feedback given by @pixeltrix

https://github.com/neerajdotname/rails/compare/6127b

I'll clean up the code a bit , add tests and will delete some code that is not needed regarding hstore etc if there is any interest in moving hstore to be on top of serialization.

[therealjasonkenney]

@neerjdotname Forgive me if this seems naive, but wouldn't it make more sense to specify using hstore in the database.yml (serialize using hstore), what would happen if that model was using sqlite instead?

[neerajsingh0101]

@bloodycelt sqlite does not support hstore. So if you are using sqlite you should not add this line in your model .

serialize :preferences, ActiveRecord::Coders::HstoreColumn.new

[prathamesh-sonpatki]

Bump. Any updates on this?

[robin850]

Isn't it related to #12395 ?

[peterc]

Pushed back another point release?

[rafaelfranca]

Yes. 😞

[wtn]

For a quick fix, I added #attr_name_will_change! before each change to hstore attributes in my application.

[etagwerker]

Bump. I just reproduced this bug in 4.0.4

I tried using #attr_name_will_change! before each change but it's not working for me.

[gdeglin]

Yikes, this is a nasty one for anyone using hstore. Just got bit by this...

Example of problematic code

def example
  player = Player.last
  player.tags["foo"] = "bar"
  puts "Initial: #{player.tags['foo']}"       # "bar"
  player.save
  puts "Reload: #{player.reload.tags['foo']}" # null (???)
  player.tags["foo"] = "bar"
  player.tags_will_change! # Workaround...
  player.save
  puts "Now?: #{player.reload.tags['foo']}" # "bar"
end
example

Output:

Initial: bar
Reload: 
Now?: bar

Version info:

ruby 2.1.0p0 (2013-12-25 revision 44422) [x86_64-darwin13.0]
Rails 4.0.4

Column info for "tags":

 tags               | hstore                      | not null default ''::hstore

[al2o3cr]

@gdeglin - I'd recommend one change to the above: player.tags_will_change! should be before the assignment. That's why it's "will" and not "did" change :)

[fabiokr]

This is a monkey patch I have been using to make that a default for hstores and arrays:

module Ext::ActiveRecord::Dirty
  private

  # Private: Overwrites the default implementation to include arrays and hstores
  # in the list of attributes that should always be saved.
  #
  # Without this, arrays and stores don't get their values updated in
  # the database.
  #
  # Relates to this issue: https://github.com/rails/rails/issues/6127
  def keys_for_partial_write
    super | self.class.columns.select do |c|
      c.try(:array) || c.type == :hstore
    end.map(&:name)
  end
end

ActiveSupport.on_load :active_record do
  include Ext::ActiveRecord::Dirty
end

[kylase]

@gdeglin I have just tested with @al2o3cr suggestion and you should execute tags_will_change! first then set tags.

[chadwtaylor]

+1 This is an issue for me too!

[senny]

This is the current expected behavior. You need to call XYZ_will_change! when modifying hstore or json attributes in place. It is related to the discussion on serialized attributes (they are always saved even when they don't change). A fix will benefit both scenarios.

@joevandyk thank you for reporting.

[saurabhnanda]

Just faced this. What is the final call:
(a) Fix serialized columns -- should be saved only when changed, OR
(b) Make sure Hstore column is always saved, whether changed or not?

[saurabhnanda]

Btw, in Postgres, hstore allows you to change the column partially, eg.

UPDATE mytable SET properties = (properties - 'deleted_key') || hstore('added_key', 'added_value')

[senny]

@saurabhnanda we keep the current behavior until we have a working solution for serialized attributes.

[chibicode]

Came here from Stack Overflow: http://stackoverflow.com/questions/20251296/how-can-i-update-a-data-records-value-with-ruby-on-rails-4-0-1-postgresql-hstor

I'm fine with the current behavior but I wish this was documented better. Does Rails 4 have a documentation page on Hstore related stuff?

[senny]

@chibicode I'm working on a PostgreSQL specific guide. See my current draft it does go into the details and will be updated in the future.

We are also working on resolving the situation with serialized attributes. Once we finished that, we can use that same behavior for json and hstore columns.

[chibicode]

@senny wow, that's very useful. Thank you!

[tilsammans]

For what it's worth, this behavior has been fixed as of #15674 and will be part of 4.2.